A gender-linked exploratory factor analysis of antisocial behavior in young adolescents.
of an accepted taxonomic approach include: (1) antisocial behaviors range from relatively mild, such as cutting classes, to severe, such as assault with a deadly weapon, yet they are often considered together; (2) some studies have focused on the acts themselves, separately classifying types of theft and drug/alcohol use, while others have focused on the person, such as violent types and runaways; and (3) diverse methods have used self-report versus the ratings of others (e.g., parents), behaviors of incarcerated delinquents as compared with antisocial behaviors in the general population, lists of behaviors with few versus many items, and samples that combine rather than separate males and females.
Research that has attempted to classify adolescent antisocial behavior in terms of logical similarity, as well as research that has focused on the individual's behavior patterns, can be used to illustrate the problems with both approaches. For example, Pfefferbaum and Wood (1994) examined three kinds of delinquency in a sample of 296 college students: interpersonal delinquency, which included serious fights and using a weapon; property delinquency, which included theft, such as stealing a car; and substance delinquency, which included drinking alcohol and using drugs. These three categories were correlated with measures of thrill-seeking, self-control, socialization, and school success. Correlations between the three classes of delinquent behaviors were found to be low to moderate: substance delinquency and property delinquency, .256; substance delinquency and interpersonal delinquency, .279; property delinquency and interpersonal delinquency, .640. Further, multiple regression analysis, with the three forms of delinquency as dependent variables, showed only one consistent predictor; namely, males showed higher scores. Otherwise, the predictors varied for each of the dependent variables.
Pfefferbaum and Wood's (1994) finding of low to moderate correlations between logically grouped classes of delinquent behavior, and differing motivations for each, provided two reasons to separate the forms of delinquency when investigating social and personality correlates. However, there may have been problems with their classification system. It was possible that stealing and drug sales shared a functional relationship, or that drug use and drug sales were unrelated because they did not share the same motivational basis. Thus, uncovering related clusters of antisocial behaviors would set the stage for an empirically based examination of personality and social correlates.
A contrasting approach has viewed antisocial behavior as unidimensional. In a study by Simons, Robertson, and Downs (1989), scores for widely varying behaviors, such as cutting classes, assault with a deadly weapon, and drug sales, were combined and then correlated with social and personality variables. If antisocial behavior is truly a unidimensional construct, then combining these behaviors was warranted, whereas if it is multidimensional, then such combination may have masked important differential associations with social and personality dimensions.
Research has also attempted to focus primarily on the individual and secondarily on the nature of the behavior itself. Moffitt (1993), taking a developmental perspective, identified two kinds of antisocial individuals: life-course-persistent offenders and adolescence-limited offenders. Life-course-persistent offenders begin engaging in antisocial acts before adolescence. In sharp contrast, adolescence-limited offenders start during adolescence and generally engage in vandalism, substance abuse, status offenses (e.g., running away from home), and theft. While the life-course-persistent offenders also engage in such behaviors, they are prone to commit more serious victim-oriented offenses. From this person-oriented theory, it was predicted that antisocial behavior would be multidimensional and that victim-oriented offenses would cluster together, as would non-victim-oriented offenses. However, Moffitt's research review and theory formulation dealt for the most part with males. Yet, recent research has indicated that the antisocial behavior of males differs from that of females in form, frequency, and variability (Rhodes & Fischer, 1993; Fergusson, Norwood, & Lynskey, 1994). This suggests that a taxonomy should be gender linked.
Empirical examinations of latent structures in self-report measures of antisocial behaviors are needed. When the construct tapped by self-report measures has been broad, and a variety of behaviors included, a unidimensional structure has been found by some researchers (Donovan & Jessor, 1985; Donovan, Jessor, & Costa, 1988) and a multidimensional structure by others (Achenbach & Edelbrock, 1991). When the construct tapped has been more limited, findings have been supportive of multidimensionality. For example, Shaw et al. (1992) used exploratory factor analysis with samples of college and high school students to investigate the factor structure of a 10-item measure of reckless behavior. A two-factor model appeared to best fit the data for both samples. Factor 1 included driving beyond the speed limit and vandalism. Factor 2 included drug use, shoplifting, and risky sexual behavior. Thorson and Powell (1987) used exploratory factor analysis to investigate a measure of "lethal" behavior and found the following four factors: a general orientation toward danger and violence; an orientation toward bravery and adventure; thrill-seeking and fast driving; and attraction to safe and unsafe activities. In comparison, Clark et al. (1990) found that reckless behavior was best described by three factors: an interest in weapons and military activities; involvement in dangerous driving and substance abuse; and smoking, drug use, and association with "bad company."
While a discussion of consistencies and inconsistencies in factor analyses of measures of reckless behavior is beyond the scope of the present research, certain generalizations can be made. First, the factor structures derived in these studies depended upon the number and variety of items employed. Second, multidimensionality of antisocial behaviors was generally the rule rather than the exception.
The current research investigated the latent structure of antisocial behaviors for early adolescent males and females. Two main research questions were investigated: Is antisocial behavior multidimensional? and Is the factor structure different for males as compared with females? Adolescents were asked whether they had ever engaged in particular antisocial behaviors and how many times they had done so in the last year. Exploratory factor analyses were performed separately by gender because differences in antisocial behavior have been noted.
The participants were 72 males and 91 females in grades 6 (n = 50), 7 (n = 50), and 8 (n = 63) of a metropolitan middle school. Approximately 88% lived in two-parent households; 79% of mothers and 78% of fathers were college graduates; 80% of the adolescents were Caucasian; and 74% of the mothers worked outside the home.
The 23-item self-report measure of antisocial behavior employed was previously used by Conger and Conger (1994) with similar-aged adolescents. The measure was originally devised for national sampling by Elliott, Huizinga, and Ageton (1985). Students were asked whether they had ever performed the particular behavior (yes or no) and the frequency of that behavior in the past year (0 = never, 1 = once, 2 = 2 to 3 times, 3 = 4 to 5 times, and 4 = 6 or more times).
Responses were subjected to separate exploratory factor analysis. For females, the alphas for yes/no and frequency data were .87 and .88, respectively. For males, alphas were .83 and .87, respectively. Validity of the measure is supported by correlations with measures of relationships with parents and siblings, value commitment, and social skills (see Simons, Robertson, & Downs, 1989; Conger & Conger, 1994; Simons, Whitbeck, Conger, & Conger, 1991).
Permission to administer the measure in classrooms was obtained from the school principal. Parents were informed that the purpose of the research was to investigate rule-breaking behaviors, and signed consent was obtained.
A t test for independent samples for males and females was conducted for all yes/no and frequency (last year) variables. As shown in Table 1, males were more likely than females to have ever taken less than $25, beaten up someone because they were angry at them, snatched a purse or wallet, broken into a place for fun, and been picked up by the police. Females were more likely than males to have ever cut classes and been drunk in a public place.
[TABULAR DATA FOR TABLE 1 OMITTED]
Data for frequency of antisocial behavior in the last year show that males, as compared with females, were more likely to have beaten up someone when they were angry, snatched a purse or wallet, destroyed property, and thrown an object to hurt or scare someone. Females, as compared with males, were more likely to have cut classes and been drunk in public places in the last year. Since about one third of the variables showed significant gender differences (t test), and analysis of variance (F) was significant at p [less than] .05 or better for about half (not shown in Table 1), factor analyses were conducted by gender.
Factor Analyses: Yes/No Data
Students' responses as to whether they had ever committed a particular act (i.e., yes/no responses) were analyzed separately by gender and were subjected to a principal component factor analysis with VARIMAX rotation. A loading of .40 was used as the cutoff for inclusion. As a general rule, single-item factors were dropped.
The results for males are shown in Table 2. Two items were omitted due to the absence of variability. Factor 8 was omitted because one of two items was shifted to another factor.
The seven-factor solution for males explained 67.8% of the total variance in antisocial behavior. Factor 1, with 24.5% of the variance, was labeled stealing/drug sales. Factor 2, with 10.8% of the variance, was called problems with alcohol. Factor 3, with 8.6% of the variance, tapped those items involving interpersonal violence with weapons. Factor 4, with 6.9% of the variance, was labeled property offenses and tapped such behaviors as destroying property, breaking into buildings, and theft of money. Factor 5, with 6.1% of the variance, appeared to have tapped thrill-seeking behaviors, such as taking a car, setting a fire for fun, and involvement with the police. Factor 6, with 5.5% of the variance, appeared to tap assertive offenses, such as fighting and sneaking into events without paying. Factor 7, with 5.4% of the variance, was labeled impact on others, and included running away and throwing objects to scare or hurt others.
Factor analysis of yes/no responses for females (see Table 3) also revealed a seven-factor solution, which explained 68.6% of the variance in antisocial behavior. Factor 1, with 27.2% of the variance, tapped violent behaviors, such as attacking someone with a weapon, throwing objects to hurt or scare others, using force to get things, and beating others up when angry. Drunkenness in public, setting fires, and sneaking into a movie or event were also associated with this factor. Factor 2, with 9.8% of the variance, appeared to involve car offenses, including drunk driving, driving without a license, and other car violations (such as speeding). Factor 3, with 8.3% of the variance, included drug sale/break-in items. Factor 4, which explained 7.0% of the variance, appeared to tap stealing, such as car theft and taking larger sums of money. Factor 5, with 5.8% of the variance, was labeled property offenses because it involved theft of smaller amounts of money and destruction [TABULAR DATA FOR TABLE 2 OMITTED] of property. Factor 6, accounting for 5.6% of the variance, appeared to involve thrill seeking, such as cutting classes, stealing, and breaking into buildings for fun. Factor 7, with 4.9% of the variance, appeared to tap escape behaviors, such as running away and being picked up by the police.
[TABULAR DATA FOR TABLE 3 OMITTED]
Factor Analyses: Frequency Data
Principal component factor analysis with VARIMAX rotation was again applied, separately by gender, using frequency of the behavior in the last year. In the case of males, two items were omitted due to lack of variability.
Table 4 Factor Loadings for Frequency (Last Year) Responses: Males (N = 72) Factor 1 2 3 4 5 Item Steal violent Steal Thrill Impact Drg Alc Seek Others Run Away -.14 .22 .44 .02 .54 Take[less than]$25 .19 .15 .69 .17 -.07 Take[greater than]$25 .57 .14 .62 .22 .09 Drive Drunk -.03 -.14 .03 .88 -.13 Cut Class .45 .78 -.03 .04 -.02 Take Car .39 -.02 .63 .49 .03 Beat Up .04 .21 .15 .59 .35 Crt Prob .10 .09 -.07 -.03 -.05 Det Jail .18 .03 -.08 -.02 .81 Snatch Wallet .02 .48 .46 .10 .23 Drunk Pub .54 .08 -.06 .06 .03 Dest Prop .27 .42 .30 .31 .46 Brk In Fun .09 .72 .14 .09 .37 Brk In Destroy .88 -.06 .19 .02 .02 Throw Obj -.03 .29 .24 .19 .57 Att Weapon .16 .64 .48 .07 .30 Sold Drug .94 .11 -.05 .01 .09 Force Weapon -.07 .70 .35 .01 -.08 Pick Police .55 .38 .33 .45 .13 Set Fire -.12 .25 .71 -.06 .18 Sneak Event .20 .40 .11 .58 .18 Eigen 6.63 2.67 1.66 1.35 1.24 Pct Var 31.6 12.7 7.9 7.0 5.8
For males, a five-factor model, presented in Table 4, explained 65.0% of the variance. Factor 1, accounting for 31.6% of the variance, was labeled stealing/drugs/alcohol. Factor 2, with 12.7% of the variance, tapped violent behavior, which included attacking someone with a weapon and using force or weapons to get things. Cutting classes, snatching purses/wallets, breaking into buildings, and damaging property were also associated with this factor. Factor 3, with 7.9% of the variance, was labeled stealing since it involved theft of money, cars, and wallets/purses. In addition, fire setting and attacking someone with a weapon were associated with this factor. Factor 4, with 7.0% of the variance, tapped thrill-seeking behavior, such as sneaking into events or movies, getting into fights, and driving drunk. Factor 5, explaining 5.8% of the variance, was labeled impact on others, and included running away, hurting or scaring others by throwing objects, and destroying property.
For females, a five-factor solution explained 67.1% of the variance in antisocial behavior (see Table 5). Factor 1, accounting for 32.9% of the variance, was labeled violent/drug/alcohol and included such items as attacking someone with a weapon, using force or weapons to get things, beating others up, selling drugs, and drunk driving. Factor 2, with 13.1% of the variance, tapped stealing, such as theft of money, destruction of property, and sneaking into movies or other events. Factor 3, with 7.5% of the variance, was labeled negative effects because the behaviors had a negative or potentially negative impact on persons or property. Factor 4, accounting for 7.2% of the variance, tapped thrill seeking and included public drunkenness and breaking into buildings for fun. Factor 5, with 6.4% of the variance, involved escape behavior, such as running away.
The present study examined the multidimensionality of antisocial behavior in young adolescent males and females. The latent structure of the antisocial behaviors in early adolescence varied depending on whether the time frame was all previous years (i.e., yes/no data) or frequency during the last year, and whether the youth was male or female.
Specifically, males, as compared with females, showed more aggressive antisocial behaviors, such as fighting when angered. Females, as compared with males, indicated greater class cutting and public drunkenness. Research by Fergusson, Norwood, and Lynskey (1994) likewise found, with a sample of 15-year-olds, that females showed greater problems with marijuana and alcohol abuse, and males demonstrated more law-breaking behavior. Rhodes and Fischer (1993) found that inner-city males were more likely to have engaged in offenses involving aggression, whereas females were more likely to have been referred to a court-related program for status offenses. The present findings supply [TABULAR DATA FOR TABLE 5 OMITTED] further evidence of greater aggressive antisocial behavior in males as compared with females (Cairns & Cairns, 1988; Offord, Boyle, & Racine, 1991).
The first three factors across all analyses, which accounted for from 43.9% to 53.5% of the variance in total antisocial behavior, as well as two thirds of the variance accounted for by the five to seven factors, involved drug/alcohol and violent behaviors. Such behaviors are clearly of considerable public concern.
For females, the first factor in both analyses was violent behavior. Violence accounted for 27.2% of the variance in antisocial behavior for all previous years and 32.9% of the variance in antisocial behavior during the last year. For both time frames, behaviors such as beating others up because they made them angry, using force or a weapon to get things, and attacking someone with a weapon were salient. Interestingly, those who committed such serious acts may also have been involved in many other antisocial behaviors. In fact, when the time frame was frequency during the last year, there were 13 separate behaviors that clustered as part of this factor. In other words, violent females are likely to display a number of other behaviors that are not specifically directed at hurting others or destroying things. The other factors in both time frame analyses were not as broad or inclusive as the violence factor, but were restricted to more specialized or limited kinds of antisocial behaviors, such as car offenses, stealing, and thrill seeking.
The factor structure of antisocial behavior for males was different from that of females in that all the factors showed greater specialization. In addition, the factor accounting for the greatest variance was not violence, as was the case for females, but stealing/drugs/alcohol or stealing/drug sales. Among males the factors were more coherent and limited. For example, stealing/drug sales involved various kinds of theft and the sale of drugs, while alcohol offenses involved driving drunk, drunkenness in public, and cutting class. The violence factor included attacking others to hurt them, using force to get things from others, and other violent behaviors, but violence was not associated with alcohol, drugs, or stealing. In other words, the males who were violent were not the same ones who were stealing and using drugs. Studies on adults have tended to find similar specialization in antisocial behavior among males (Brennan, Mednick, & Kandel, 1991).
Additional support for specialization comes from longitudinal research in developmental psychology. Aggressive behavior for males shows continuity from childhood to adulthood (Farrington, 1991; Olweus, 1979), and the relation between early aggressive behavior and later violent criminal behavior for males is positive (Farrington, 1991). Further, Moffitt (1993) has theorized that specialization in victim offenses, such as violence and fraud, is likely to be characteristic of a qualitatively distinct subset of individuals called life-course-persistent offenders. However, this theory is based, for the most part, on males. The present research supports a taxonomy of adolescent antisocial behavior that is, in addition, gender sensitive.
This investigation represents an important step in the empirical classification of antisocial behavior in early adolescence. It is not known whether research on middle and late adolescents would reveal similar factors. In addition, the participants in this study came primarily from intact families, and it is not known whether a more diverse sample with more serious antisocial behaviors would produce similar results. Further, the research on reckless behavior cited earlier would suggest that latent structures will vary according to the variety and number of behaviors included. The present study suggests that gender and time frame influence these structures as well. Further research might investigate whether females specialize somewhat later in adolescence, and whether males continue to specialize. While self-report data have been found to be valid and reliable (Elliott, Huizinga, & Morse, 1986), future research might look into whether rater (i.e., self, parent, teacher) convergence provides greater confidence in derived factors, as has been done with adolescent problem behaviors (Achenbach & Edelbrock, 1991). Once a firm, empirically based taxonomy of antisocial behaviors has been established, the search for personality and social correlates will likely have a more secure foundation.
Achenbach, T., & Edelbrock, C. (1991). Manual for the Child Behavior Checklist and Revised Child Behavior Profile. Burlington, VT: authors.
Brennan, P., Mednick, S. T., & Kandel, E. (1991). Congenital determinants of violent and property offending. In D. Peppler & K. Rubin (Eds.), The development and treatment of childhood aggression (pp. 81-92). Hillsdale, NJ: Erlbaum.
Cairns, R. B., & Cairns, B. D. (1988). The sociogenesis of self concepts. In N. Bolger, A. Caspi, G. Downey, & M. Moorehouse (Eds.), Persons in social context: Developmental processes (pp. 181-202). New York: Cambridge University Press.
Clark, D. C., Sommerfeldt, L., Schwartz, M., Hedeker, D., & Watel, L. (1990). Physical recklessness in adolescence: Trait or by-product of depressive/suicidal states? Journal of Nervous and Mental Disorders, 178, 423-433.
Conger, K. J., & Conger, R. D. (1994). Differential parenting and change in sibling differences in delinquency. Journal of Family Psychology, 8, 287-302.
Dodge, K. (1991). The structure and function of reactive and proactive aggression. In D. Peppler & K. Rubin (Eds.), The development and treatment of childhood aggression (pp. 201-218). Hillsdale, NJ: Erlbaum.
Donovan, J. E., & Jessor, R. (1985). Structure of problem behavior in adolescence and young adulthood. Journal of Consulting and Clinical Psychology, 53, 890-904.
Donovan, J. E., Jessor, R., & Costa, F. M. (1988). Syndrome of problem behaviors in adolescence: A replication. Journal of Consulting and Clinical Psychology, 56, 762-765.
Elliott, D. S., Huizinga, D., & Ageton, S. S. (1985). Explaining delinquency and drug use. Beverly Hills, CA: Sage.
Elliott, D. S., Huizinga, D., & Morse, B. (1986). Self-reported violent offending: A descriptive analysis of juvenile violent offenders and their offending careers. Journal of Interpersonal Violence, 1, 472-514.
Farrington, D. (1991). Childhood aggression and adult violence: Early precursors and later life outcome. In D. Peppler & K. Rubin (Eds.), The development and treatment of childhood aggression (pp. 5-29). Hillsdale, NJ: Erlbaum.
Fergusson, D., Norwood, L., & Lynskey, M. (1994). The comorbidities of adolescent problem behaviors: A latent class model. Journal of Abnormal Child Psychology, 22, 339-354.
Moffitt, T. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100, 674-701.
Offord, D., Boyle, M., & Racine, Y. (1991). The epidemiology of antisocial behavior in childhood and adolescence. In D. Peppler & K. Rubin (Eds.), The development and treatment of childhood aggression (pp. 31-54). Hillsdale, NJ: Erlbaum.
Olweus, D. (1979). Stability of aggressive reaction patterns in males: A review. Psychological Bulletin, 86, 852-875.
Pfefferbaum, B., & Wood, P. B. (1994). Self-report study of implusive and delinquent behavior in college students. Journal of Adolescence, 5, 295-302.
Rhodes, J. E., & Fischer, K. (1993). Spanning the gender gap: Gender differences in delinquency among inner-city adolescents. Adolescence, 28(112), 879-889.
Shaw, D. S., Wagner, E. F., Arnett, J., & Aber, M. S. (1992). The factor structure of the Reckless Behavior Questionnaire. Journal of Youth and Adolescence, 21, 305-323.
Simons, R., Robertson, J. F., & Downs, W. R. (1989). The nature of the association between parental rejection and delinquent behavior. Journal of Youth and Adolescence, 18, 297-322.
Simons, R., Whitbeck, L., Conger, R., & Conger, K. (1991). Parenting factors, social skills, and value commitments as precursors to school failure, involvement with deviant peers and delinquent behavior. Journal of Youth and Adolescence, 20, 645-664.
Thorson, J. A., & Powell, F. C. (1987). Factor structure of the Lethal Behavior Scale. Psychological Reports, 61, 807-810.
Watts, W. D., & Wright, L. S. (1990). The relationship of alcohol, tobacco, marijuana and other illegal drug use to delinquency among Mexican-American, Black, and White adolescent males. Adolescence, 25(97), 171-181.